Spark SQL Guide

CSV Files

Spark SQL provides spark.read().csv("file_name") to read a file or directory of files in CSV format into Spark DataFrame, and dataframe.write().csv("path") to write to a CSV file. Function option() can be used to customize the behavior of reading or writing, such as controlling behavior of the header, delimiter character, character set, and so on.

# spark is from the previous example
sc = spark.sparkContext

# A CSV dataset is pointed to by path.
# The path can be either a single CSV file or a directory of CSV files
path = "examples/src/main/resources/people.csv"

df = spark.read.csv(path)
df.show()
# +------------------+
# |               _c0|
# +------------------+
# |      name;age;job|
# |Jorge;30;Developer|
# |  Bob;32;Developer|
# +------------------+

# Read a csv with delimiter, the default delimiter is ","
df2 = spark.read.option("delimiter", ";").csv(path)
df2.show()
# +-----+---+---------+
# |  _c0|_c1|      _c2|
# +-----+---+---------+
# | name|age|      job|
# |Jorge| 30|Developer|
# |  Bob| 32|Developer|
# +-----+---+---------+

# Read a csv with delimiter and a header
df3 = spark.read.option("delimiter", ";").option("header", True).csv(path)
df3.show()
# +-----+---+---------+
# | name|age|      job|
# +-----+---+---------+
# |Jorge| 30|Developer|
# |  Bob| 32|Developer|
# +-----+---+---------+

# You can also use options() to use multiple options
df4 = spark.read.options(delimiter=";", header=True).csv(path)

# "output" is a folder which contains multiple csv files and a _SUCCESS file.
df3.write.csv("output")

# Read all files in a folder, please make sure only CSV files should present in the folder.
folderPath = "examples/src/main/resources"
df5 = spark.read.csv(folderPath)
df5.show()
# Wrong schema because non-CSV files are read
# +-----------+
# |        _c0|
# +-----------+
# |238val_238|
# |  86val_86|
# |311val_311|
# |  27val_27|
# |165val_165|
# +-----------+
Find full example code at "examples/src/main/python/sql/datasource.py" in the Spark repo.
// A CSV dataset is pointed to by path.
// The path can be either a single CSV file or a directory of CSV files
val path = "examples/src/main/resources/people.csv"

val df = spark.read.csv(path)
df.show()
// +------------------+
// |               _c0|
// +------------------+
// |      name;age;job|
// |Jorge;30;Developer|
// |  Bob;32;Developer|
// +------------------+

// Read a csv with delimiter, the default delimiter is ","
val df2 = spark.read.option("delimiter", ";").csv(path)
df2.show()
// +-----+---+---------+
// |  _c0|_c1|      _c2|
// +-----+---+---------+
// | name|age|      job|
// |Jorge| 30|Developer|
// |  Bob| 32|Developer|
// +-----+---+---------+

// Read a csv with delimiter and a header
val df3 = spark.read.option("delimiter", ";").option("header", "true").csv(path)
df3.show()
// +-----+---+---------+
// | name|age|      job|
// +-----+---+---------+
// |Jorge| 30|Developer|
// |  Bob| 32|Developer|
// +-----+---+---------+

// You can also use options() to use multiple options
val df4 = spark.read.options(Map("delimiter" -> ";", "header" -> "true")).csv(path)

// "output" is a folder which contains multiple csv files and a _SUCCESS file.
df3.write.csv("output")

// Read all files in a folder, please make sure only CSV files should present in the folder.
val folderPath = "examples/src/main/resources";
val df5 = spark.read.csv(folderPath);
df5.show();
// Wrong schema because non-CSV files are read
// +-----------+
// |        _c0|
// +-----------+
// |238val_238|
// |  86val_86|
// |311val_311|
// |  27val_27|
// |165val_165|
// +-----------+
Find full example code at "examples/src/main/scala/org/apache/spark/examples/sql/SQLDataSourceExample.scala" in the Spark repo.
import org.apache.spark.sql.Dataset;
import org.apache.spark.sql.Row;

// A CSV dataset is pointed to by path.
// The path can be either a single CSV file or a directory of CSV files
String path = "examples/src/main/resources/people.csv";

Dataset<Row> df = spark.read().csv(path);
df.show();
// +------------------+
// |               _c0|
// +------------------+
// |      name;age;job|
// |Jorge;30;Developer|
// |  Bob;32;Developer|
// +------------------+

// Read a csv with delimiter, the default delimiter is ","
Dataset<Row> df2 = spark.read().option("delimiter", ";").csv(path);
df2.show();
// +-----+---+---------+
// |  _c0|_c1|      _c2|
// +-----+---+---------+
// | name|age|      job|
// |Jorge| 30|Developer|
// |  Bob| 32|Developer|
// +-----+---+---------+

// Read a csv with delimiter and a header
Dataset<Row> df3 = spark.read().option("delimiter", ";").option("header", "true").csv(path);
df3.show();
// +-----+---+---------+
// | name|age|      job|
// +-----+---+---------+
// |Jorge| 30|Developer|
// |  Bob| 32|Developer|
// +-----+---+---------+

// You can also use options() to use multiple options
java.util.Map<String, String> optionsMap = new java.util.HashMap<String, String>();
optionsMap.put("delimiter",";");
optionsMap.put("header","true");
Dataset<Row> df4 = spark.read().options(optionsMap).csv(path);

// "output" is a folder which contains multiple csv files and a _SUCCESS file.
df3.write().csv("output");

// Read all files in a folder, please make sure only CSV files should present in the folder.
String folderPath = "examples/src/main/resources";
Dataset<Row> df5 = spark.read().csv(folderPath);
df5.show();
// Wrong schema because non-CSV files are read
// +-----------+
// |        _c0|
// +-----------+
// |238val_238|
// |  86val_86|
// |311val_311|
// |  27val_27|
// |165val_165|
// +-----------+
Find full example code at "examples/src/main/java/org/apache/spark/examples/sql/JavaSQLDataSourceExample.java" in the Spark repo.

Data Source Option

Data source options of CSV can be set via:

Property NameDefaultMeaningScope
sep
delimiter		 , Sets a separator for each field and value. This separator can be one or more characters. read/write
extension csv Sets the file extension for the output files. Must be non-empty and contain only letters. If it matches a compression codec suffix (for example, gz or zst), set compression too; otherwise Spark may try to decompress plain-text output when reading it back. write
encoding
charset		 UTF-8 For reading, decodes the CSV files by the given encoding type. For writing, specifies encoding (charset) of saved CSV files. CSV built-in functions ignore this option. read/write
quote " Sets a single character used for escaping quoted values where the separator can be part of the value. For reading, if you would like to turn off quotations, you need to set not null but an empty string. For writing, if an empty string is set, it uses u0000 (null character). read/write
quoteAll false A flag indicating whether all values should always be enclosed in quotes. Default is to only escape values containing a quote character. write
escape \ Sets a single character used for escaping quotes inside an already quoted value. read/write
escapeQuotes true A flag indicating whether values containing quotes should always be enclosed in quotes. Default is to escape all values containing a quote character. write
comment Sets a single character used for skipping lines beginning with this character. By default, it is disabled. read
header false For reading, uses the first line as names of columns. For writing, writes the names of columns as the first line. Note that if the given path is a RDD of Strings, this header option will remove all lines same with the header if exists. CSV built-in functions ignore this option. read/write
inferSchema false Infers the input schema automatically from data. It requires one extra pass over the data. CSV built-in functions ignore this option, except as noted under variantRespectInferSchema. read
preferDate true During schema inference (inferSchema), attempts to infer string columns that contain dates as Date if the values satisfy the dateFormat option or default date format. For columns that contain a mixture of dates and timestamps, try inferring them as TimestampType if timestamp format not specified, otherwise infer them as StringType. read
enforceSchema true If it is set to true, the specified or inferred schema will be forcibly applied to datasource files, and headers in CSV files will be ignored. If the option is set to false, the schema will be validated against all headers in CSV files in the case when the header option is set to true. Field names in the schema and column names in CSV headers are checked by their positions taking into account spark.sql.caseSensitive. Though the default value is true, it is recommended to disable the enforceSchema option to avoid incorrect results. CSV built-in functions ignore this option. read
ignoreLeadingWhiteSpace false (for reading), true (for writing) A flag indicating whether or not leading whitespaces from values being read/written should be skipped. read/write
ignoreTrailingWhiteSpace false (for reading), true (for writing) A flag indicating whether or not trailing whitespaces from values being read/written should be skipped. read/write
nullValue Sets the string representation of a null value. Since 2.0.1, this nullValue param applies to all supported types including the string type. read/write
treatNullAsEmptyString Controls how null values are written when nullValue is left at its default (empty string). When true, a null is written through emptyValue (a quoted empty string "" by default), which makes it indistinguishable from an actual empty string. When false, a null is written as a bare, unquoted empty token, so it can be told apart from an empty string. When unset, the write follows the session default. Setting a non-empty nullValue makes this option a no-op, since the nullValue is then written verbatim. This option only affects writing. write
nanValue NaN Sets the string representation of a non-number value. read
positiveInf Inf Sets the string representation of a positive infinity value. read
negativeInf -Inf Sets the string representation of a negative infinity value. read
dateFormat yyyy-MM-dd Sets the string that indicates a date format. Custom date formats follow the formats at Datetime Patterns. This applies to date type. read/write
timestampFormat yyyy-MM-dd'T'HH:mm:ss[.SSS][XXX] Sets the string that indicates a timestamp format. Custom date formats follow the formats at Datetime Patterns. This applies to timestamp type. read/write
timestampNTZFormat yyyy-MM-dd'T'HH:mm:ss[.SSS] Sets the string that indicates a timestamp without timezone format. Custom date formats follow the formats at Datetime Patterns. This applies to timestamp without timezone type, note that zone-offset and time-zone components are not supported when writing or reading this data type. read/write
timeFormat HH:mm:ss Sets the string that indicates a time format. Custom time formats follow the formats at Datetime Patterns. This applies to time type. read/write
enableDateTimeParsingFallback Enabled if the time parser policy has legacy settings or if no custom date or timestamp pattern was provided. Allows falling back to the backward compatible (Spark 1.x and 2.0) behavior of parsing dates and timestamps if values do not match the set patterns. read
maxColumns 20480 Defines a hard limit of how many columns a record can have. read
maxCharsPerColumn -1 Defines the maximum number of characters allowed for any given value being read. By default, it is -1 meaning unlimited length read
mode PERMISSIVE Allows a mode for dealing with corrupt records during parsing. It supports the following case-insensitive modes. Note that Spark tries to parse only required columns in CSV under column pruning. Therefore, corrupt records can be different based on required set of fields. This behavior can be controlled by spark.sql.csv.parser.columnPruning.enabled (enabled by default). In particular, when multiLine is disabled, a quoted value containing a line separator (lineSep) splits one source record into two before mode is applied. Each of the two is then evaluated on its own, so a record whose token count happens to match the schema is retained as valid even though its value was truncated. This holds under DROPMALFORMED as well, since that mode drops only the records it finds malformed. An action requiring no columns (a bare count(), for example) may surface none of this because of column pruning.
  • PERMISSIVE: when it meets a corrupted record, puts the malformed string into a field configured by columnNameOfCorruptRecord, and sets malformed fields to null. To capture the malformed string, a user can set a string type field named columnNameOfCorruptRecord in a user-defined schema. If a schema does not have the field, the corrupt record is still retained with its malformed fields set to null, but the malformed string is not available. A record with a different number of tokens than the schema is a corrupted record, and is handled by this mode like any other; it is still parsed as far as it can be, with null set for tokens the record does not have and extra tokens dropped.
  • DROPMALFORMED: ignores the whole corrupted records. This mode is unsupported in the CSV built-in functions.
  • FAILFAST: throws an exception when it meets corrupted records.
 read
columnNameOfCorruptRecord (value of spark.sql.columnNameOfCorruptRecord configuration) Allows renaming the new field having malformed string created by PERMISSIVE mode. This overrides spark.sql.columnNameOfCorruptRecord. read
singleVariantColumn (none) If specified, the entire CSV record is parsed and stored as a single column of VariantType with the given column name, instead of being split into individual fields. By default, scalar values are type-inferred inside the variant on a best-effort basis (for example, "0001" may be inferred as the integer 1). To preserve scalar values as strings, set variantRespectInferSchema to true and inferSchema to false. read
variantRespectInferSchema false When reading into a VARIANT (see singleVariantColumn or a VariantType column in the schema), controls whether the inferSchema option is honored. When true and inferSchema is false, scalar values are preserved as strings inside the variant instead of being inferred as boolean, long, decimal, date, or timestamp. When false (the default), scalar types are inferred on a best-effort basis regardless of inferSchema. Enabling this option forces from_csv to observe inferSchema when reading into a VARIANT. read
multiLine false Allows a row to span multiple lines, by parsing line breaks within quoted values as part of the value itself. CSV built-in functions ignore this option.
When this option is disabled (the default), a line break inside a quoted value terminates the record at that break: the value is truncated, and the rest of the value begins a new record. Either or both of the resulting records may then be malformed (for example when the token count no longer matches the schema); how they are handled is controlled by mode.		 read
charToEscapeQuoteEscaping escape or \0 Sets a single character used for escaping the escape for the quote character. The default value is escape character when escape and quote characters are different, \0 otherwise. read/write
samplingRatio 1.0 Defines fraction of rows used for schema inferring. CSV built-in functions ignore this option. read
emptyValue (for reading), "" (for writing) Sets the string representation of an empty value. read/write
locale en-US Sets a locale as language tag in IETF BCP 47 format. For instance, this is used while parsing dates and timestamps. read
lineSep \r, \r\n and \n (for reading), \n (for writing) Defines the line separator that should be used for parsing/writing. Maximum length is 1 character. CSV built-in functions ignore this option. read/write
unescapedQuoteHandling STOP_AT_DELIMITER Defines how the CsvParser will handle values with unescaped quotes.
  • STOP_AT_CLOSING_QUOTE: If unescaped quotes are found in the input, accumulate the quote character and proceed parsing the value as a quoted value, until a closing quote is found.
  • BACK_TO_DELIMITER: If unescaped quotes are found in the input, consider the value as an unquoted value. This will make the parser accumulate all characters of the current parsed value until the delimiter is found. If no delimiter is found in the value, the parser will continue accumulating characters from the input until a delimiter or line ending is found.
  • STOP_AT_DELIMITER: If unescaped quotes are found in the input, consider the value as an unquoted value. This will make the parser accumulate all characters until the delimiter or a line ending is found in the input.
  • SKIP_VALUE: If unescaped quotes are found in the input, the content parsed for the given value will be skipped and the value set in nullValue will be produced instead.
  • RAISE_ERROR: If unescaped quotes are found in the input, a TextParsingException will be thrown.
 read
compression
codec		 (none) Compression codec to use when saving to file. This can be one of the known case-insensitive shorten names (none, bzip2, gzip, lz4, snappy and deflate). CSV built-in functions ignore this option. write
timeZone (value of spark.sql.session.timeZone configuration) Sets the string that indicates a time zone ID to be used to format timestamps in the JSON datasources or partition values. The following formats of timeZone are supported:
  • Region-based zone ID: It should have the form 'area/city', such as 'America/Los_Angeles'.
  • Zone offset: It should be in the format '(+|-)HH:mm', for example '-08:00' or '+01:00'. Also 'UTC' and 'Z' are supported as aliases of '+00:00'.
Other short names like 'CST' are not recommended to use because they can be ambiguous. 		read/write

Other generic options can be found in Generic File Source Options.